Hierarchical object detection and tracking with an Implicit Shape Model

One important field in machine vision is object tracking. In most real-world applications, multiple object instances of different classes which influence each others appearance are of interest. Independent treatment of these objects in detection and tracking is not sufficient. In this paper, we present an object tracking algorithm which takes this into account and builds on an Implicit Shape Model (ISM) based detection and tracking. This suits it for the task of tracking multiple interacting objects in a cascaded approach because the local feature based ISM approach allows inference of the image regions classification was based on and hence allows for differentiation of the objects on the sensory level. We conduct two experiments: The first experiment demonstrates tracking on a surveillance dataset for the case of persons. In the second experiment, hierarchical tracking is performed for the cases of persons on ships and bags which are carried by persons.